[rogeralsing]

Summary

• drop assistant-message payloads that retain the default key identifier to keep virtual agent debug chatter out of the web transcript
• cover the behaviour with a unit test and document the filtering in the backend context

Testing

• node --experimental-vm-modules node_modules/jest/bin/jest.js packages/web/backend/src/server/tests/utils.test.ts --runTestsByPath

---

https://chatgpt.com/codex/tasks/task_e_68fb975dcb2c83288a8f13200e098b85

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Your team has set up Codex to review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

Codex can also answer questions or update the PR. Try commenting "@codex address that feedback".

[chatgpt-codex-connector]

Avoid filtering all assistant messages by default

The new guard drops any assistant-message whose __id starts with "key". The core runtime emits IDs with the 'key' prefix by default (packages/core/src/agent/loop.ts sets idPrefix = 'key', and the web backend never overrides it when creating the WebSocket binding), so every assistant message produced by the main agent now satisfies this condition and is discarded. In the web backend this means no assistant responses reach the browser at all, not just the debug streams this change intended to hide. Consider tightening the filter to the specific debug identifiers rather than the default prefix.

Useful? React with 👍 / 👎.